Definition
Adaptive RAG is a dynamic architectural pattern that utilizes a classifier or routing layer to determine the optimal retrieval strategy based on the complexity and intent of a user query. It balances system efficiency and accuracy by choosing between direct LLM generation, standard RAG, or iterative multi-hop retrieval, effectively trading a small routing latency for significant gains in response quality and cost optimization.
Unlike static RAG, which treats every query with the same retrieval logic, Adaptive RAG changes the workflow based on the query's difficulty.
"A Triage Nurse in an Emergency Room directing patients to a quick bandage, a general practitioner, or a specialist surgeon based on the severity of their symptoms."
- Query Routing(Prerequisite)
- Self-RAG(Complementary Technique)
- Multi-hop Retrieval(Component Strategy)
- Corrective RAG (CRAG)(Component Strategy)
Conceptual Overview
Adaptive RAG is a dynamic architectural pattern that utilizes a classifier or routing layer to determine the optimal retrieval strategy based on the complexity and intent of a user query. It balances system efficiency and accuracy by choosing between direct LLM generation, standard RAG, or iterative multi-hop retrieval, effectively trading a small routing latency for significant gains in response quality and cost optimization.
Disambiguation
Unlike static RAG, which treats every query with the same retrieval logic, Adaptive RAG changes the workflow based on the query's difficulty.
Visual Analog
A Triage Nurse in an Emergency Room directing patients to a quick bandage, a general practitioner, or a specialist surgeon based on the severity of their symptoms.